Supported services¶

common.ai reaches external services on two axes: the model providers its connections talk to, and the systems its toolsets talk to. For per-model detail beyond the vendor level (exact model ids, pricing, capabilities), see pydantic-ai’s model list.

This page is the complete list of what this provider reaches; for a curated, install- and credential-oriented walkthrough of each model vendor, see Supported model providers.

Connections¶

Connection type	External services
Pydantic AI	Alibaba Cloud Model Studio, Anthropic, AWS Bedrock, AWS Bedrock Mantle, Azure OpenAI, Cerebras, Cohere, Crusoe, DeepSeek, Fireworks AI, GitHub Copilot, Google Gemini, Google Vertex AI, Groq, Heroku, Hugging Face, LiteLLM, Mistral AI, Moonshot AI, Nebius, Ollama, OpenAI, OpenRouter, OVHcloud, Pydantic AI Gateway, SambaNova, Snowflake Cortex, Together AI, TypeSafe, Vercel AI Gateway, vLLM, xAI, Z.AI
Pydantic AI (Azure OpenAI)	Azure OpenAI
Pydantic AI (AWS Bedrock)	AWS Bedrock
Pydantic AI (Google Vertex AI)	Google Vertex AI
MCP Server	Any MCP server (user-supplied endpoint)
LangChain	OpenAI, Anthropic, Groq, Mistral AI, DeepSeek, Ollama, vLLM
LlamaIndex	OpenAI

Toolsets¶

Toolset	External services
HookToolset	Any Airflow connection, through its provider hook
SQLToolset	Any common.sql database connection (with table allowlists and AST validation)
DataFusionToolset	Object storage (S3, local filesystem, Apache Iceberg) via Apache DataFusion
LoggingToolset	Airflow task logs
MCPToolset	Any MCP server (user-supplied endpoint)
ObjectStorageToolset	Object storage (S3, GCS, Azure Blob Storage, or any store `ObjectStoragePath` can open), read-only
SandboxToolset	Docker Sandboxes (the shipped sbx backend; other backends can be added via SandboxBackend)
AgentSkillsToolset	Agent skills from local paths and remote Git repositories
LangChain Bridge	LangChain tools
Managed Agent Toolsets	Snowflake Cortex Agents; Amazon Bedrock AgentCore; Azure AI Foundry; Vertex AI Agent Engine

Notes¶

The langchain connection type’s row above is limited to OpenAI-compatible credential surfaces (api_key + optional base_url); see Supported providers for the providers that reject those kwargs and are not usable through it.
Azure OpenAI, Google Vertex AI, and AWS Bedrock are also reachable through the generic pydanticai connection type, but each has its own dedicated connection type (Pydantic AI (Azure OpenAI) connection, Pydantic AI (Google Vertex AI) connection, Pydantic AI (AWS Bedrock) connection) for their non-standard auth.
Most model providers need an extra installed alongside apache-airflow-providers-common-ai — see the “Choosing extras” section of Installation.
“Pydantic AI Gateway” in the pydanticai row is a routing layer, not an upstream vendor in its own right: set the connection’s Model field to gateway/<vendor>:<model> (for example gateway/anthropic:claude-sonnet-5) to send the request through it instead of directly to the vendor. It currently routes to Anthropic, AWS Bedrock, Google Gemini, Google Vertex AI, Groq, and OpenAI — all already listed above in their own right.